Compare
Mem0 extracts and stores memories about each user of your application. Wordcell keeps your own working memory in Markdown files you own.
Mem0 distills the messages you send into stored facts, managed or self-hosted. Wordcell notes hold what a person or agent chose to write.
Mem0’s features and license were checked on September 26, 2026.
How they differ
Each Mem0 entry links the page it comes from.
| Topic | Mem0 | Wordcell |
|---|---|---|
| Where memory lives | The Platform is managed: Mem0 runs the vector store, language model, and embedder. The open-source SDK runs them on infrastructure you provision. | Markdown files in a folder you choose, usually a Git repository you already have. |
| How memories form | Your application sends messages to add; a model extracts durable facts, deduplicates them, and links the entities they mention. | You or your agent write notes and typed relations, and you review them in Git like code. |
| Whose memory | Memories are scoped by user_id, agent_id, or run_id, and by app_id on the Platform, so each end user of your product keeps separate memories. | One person’s or one team’s vault. It does not store memory for the users of your product. |
| Platform and open source | Several features are Platform-only: the entity graph, temporal reasoning, memory decay, webhooks, and memory export. The open-source SDK is Apache-2.0. | One MIT-licensed codebase. There is no hosted tier, so no feature list is split across one. |
| Change over time | On the Platform, temporal reasoning boosts memories whose dates match the query’s time, and memory decay dampens stale ones at search time. | A supersedes relation points from a newer note to the one it replaces, and Git keeps every earlier version. |
| Cost | The SDK is free under Apache-2.0, and you pay for the vector store, LLM, and embedder you provision. The Platform is the managed service: an API key is all you configure. | Free and MIT licensed. Reranking is optional; it sends the query and each candidate’s title, path, and up to 512 bytes of its snippet to a paid provider. |
| Agent access | Python and TypeScript SDKs, a REST API, and a hosted MCP server at mcp.mem0.ai behind OAuth or an API key. | wordcell mcp serves a vault to local MCP clients over standard input and output. It is available from source until the next release. |
Which to choose
The deciding question is whose memory you keep: the users of your product, or yourself and your agents.
Choose Mem0 when
- Your product should remember facts about each end user without anyone writing notes.
- You want extraction, storage, and retrieval managed for you, with webhooks and memory export on top.
- You are prepared to provision a vector store, a language model, and an embedder for the open-source SDK.
Choose Wordcell when
- Your agents keep memory about your own work, and you want it in files you can read, edit, and diff.
- You want every memory change reviewed in Git, like code.
- You want memory as plain files, with no hosted service, no account, and no usage bill.
Read the full comparison with Supermemory, Zep, and other tools
Moving memories
Mem0 memories are extracted facts behind an API, not files you already have. There is no importer.
- Export what exists first: the Platform’s memory export runs structured jobs over filtered memories, and the open-source SDK returns them from
get_all. - Decide which facts deserve a note. A Wordcell note holds what a person or agent chose to write, so author it rather than convert it.
- There is no
wordcell import mem0and no/migratepage; moving means writing the notes.
Evidence
Wordcell has published no head-to-head comparison with Mem0 of retrieval quality or speed.
Mem0 publishes its own figures on its research page. The benchmarks page shows them with their sources, as published figures rather than a matched ranking.